Fig. 3

Precision and recall in relation to drug selection and ranking criteria. a, b Precision and recall at different z _c cut-offs. In most cell types, precision increased with decreasing z _c. Too stringent z _c cut-offs, by contrast, led to the exclusion of almost all candidates, including approved RA drugs. c, d Precision and recall after stepwise application of drug selection criteria. Bars with no pattern represent a z _c cut-off that Guney et al.¹⁰ had previously found to offer good coverage of known drug-disease pairs. The dotted pattern represents precision at z _c <  − 1.64 (corresponding to one-sided P < 0.05). The striped pattern represents precision among candidates with z _c <  − 1.64 and d _c < 1 (in other words, candidates passing network proximity-based selection). The crosshatch pattern represents candidates passing network proximity and pharmacological action selection. The application of selection criteria substantially increased the precision among central cell types. e Precision for approved RA drugs for which clinical trials had been registered for RA, and literature evidence (only for the top 100 ranking drugs) among candidates. Drug ranking included rank ties. Literature evidence was gathered for the top 100 ranked candidates and is presented as triangles with a dashed line. The colour legend for (a–d) is depicted in (f)